The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests data from GitHub issues and interpolates it into sub-agent prompts, creating a surface for indirect prompt injection. This is a standard risk for LLM-based tools and is mitigated here by specific role-based instructions and restricted file-writing tasks.
Ingestion points: GitHub issue title and body accessed via gh issue view in SKILL.md.
Boundary markers: The prompt uses labels like IDEA: and RESEARCH SUMMARY: to delineate ingested content for the sub-agents.
Capability inventory: Sub-agents are instructed to write only to specific documentation files; the main agent handles the Git and GitHub CLI interactions.
Sanitization: While the issue body is not sanitized, the idea-slug used for file paths and branch names is derived by lowercasing and removing special characters from the title.
[COMMAND_EXECUTION]: The skill executes shell commands for git and gh operations. It uses a sanitization step to convert issue titles into 'slugs' before they are used in file paths or branch names, which prevents common injection or path traversal vectors via the issue title.

planner